NVIDIA RTX 40 series is coming, not surprisingly will use TSMC N4 4nm process, in technology than Samsung N8 8nm too strong. Authoritative bloggers kopite7kimi exposed the latest news, RTX 40 series of large core AD102, to be used in RTX 4090, RTX 4090 Ti, will integrate up to 75 billion transistors!
This is almost comparable to the GH100 core for accelerated computing, the latter is also TSMC 4nm process, integrated up to 80 billion transistors, AD102 is only a mere 5 billion short.
In addition, RTX 30 series large core GA102 integrated 28.3 billion transistors, AD102 is as much as 2.65 times it, is the largest increase in recent years.
Further ahead, 12nm TU102 18.6 billion, 12nm GV100 21.1 billion, 16nm GP102 12 billion, their densities are 25 million per square millimeter, GA102 significantly increased to 45.1 million per square millimeter.
AD102 core integrated 18,432 CUDA cores, compared to GA102 increased by about 71%, the memory specifications are basically the same, I do not know where so many transistors are used. Is there a secret weapon?
Now the suspense depends on how large the area of the AD102 core, how high the degree of integration.
